dueling spam filters?

[mikelcu]

As I understand it, spam information is not stored per-user. Given this, how does the filtering handle it when a mail is delivered to more than one user, and one of the users marks it as spam while the other does not? Is there a spam/ham conflict here?

The situation I have is that I and several others receive mail via a distro list. There are a couple notorious spammers that hit this list, and though I've marked their mails as spam dozens of times, the spam score on these mails never seems to go up and they always make it to my inbox. I don't think the other recipients are marking any of the same mails as spam, which is why I'm wondering if we're canceling each other out...

-M

[uxbod]

The SA Bayes is tokenized so it is possible users are cancelling the effects out ... What I would say is what AS systems do you have in place eg. additional SA rules, SaneSecurity sigs, RBLs, Greylisting etc

[mikelcu]

We have RBL filtering on a few MTAs aside from Zimbra, but the only "smart" filters are on our ZCS server at the moment. We're already looking into adding more and better filters outside ZCS, but I'm mostly curious about why SA would still be allowing these mails through despite the repeated spam-marking.

Assuming that there is the potential for conflicting tokens inside SA, under what circumstance would that happen? If one recipient marks a mail as spam and the others do nothing, could that do it? Or, would one user have to mark something as spam, and another user explicitly mark the same mail as not-junk?

I guess the question is how does the Zimbra ham training work? Does it only respond to false positives marked as not-junk, or can a user's passive acceptance of one mail in some way taint the spam training?

-M